Download as pdf or txt
Download as pdf or txt
You are on page 1of 11

Composite Structures 154 (2016) 527–537

Contents lists available at ScienceDirect

Composite Structures
journal homepage: www.elsevier.com/locate/compstruct

Economic-design optimization of cross laminated timber plates with ribs


Andjelka Stanić, Blaž Hudobivnik, Boštjan Brank ⇑
University of Ljubljana, Faculty of Civil and Geodetic Engineering, Jamova c. 2, 1000 Ljubljana, Slovenia

a r t i c l e i n f o a b s t r a c t

Article history: An economic-design optimization of cross-laminated timber (CLT) plate with stiffening ribs is presented.
Received 27 April 2016 For the structural analysis, an enhanced assumed strain (EAS) solid finite element is used. It behaves well
Revised 6 July 2016 for thin plates (with no shear locking) and delivers reasonable approximations for the transverse shear
Accepted 25 July 2016
stresses in layered composites. Eurocodes 5 (EC5) are followed in defining the optimization constraints,
Available online 28 July 2016
which include deflections, stresses and fundamental eigenfrequency. The gradient optimization is per-
formed. Analytical expressions for sensitivities are obtained by an automatic differentiation tool. The
Keywords:
result is an economic timber plate configuration that complies with the EC5 requirements. Numerical
Cross laminated timber
Ribbed timber plate
examples are presented in order to illustrate the approach.
Economic-design optimization Ó 2016 Elsevier Ltd. All rights reserved.

1. Introduction important issue in this respect is complete understanding of the


mechanism taking place in wood materials as well as CLT failure
The cross laminated timber (CLT) is relatively new building mechanisms. Another important issue is optimal use of wood
material with a lot of potential in the timber construction industry material in timber structures, which is the topic addressed in this
[1]. Its economic design can support the developments related to work. We propose an algorithm for economic-design of CLT plates
this high-performance wood product and strengthen its competi- with or without timber ribs.
tiveness towards other building materials. From the material point
of view, CLT is highly anisotropic and shear compliant laminated 2. Optimization of ribbed CLT plate
composite. In multi-storey timber buildings it is used for structural
plates. It can be applied for plates with larger spans if stiffened by 2.1. Problem description
timber ribs.
The design of CLT plates is based on design recommendations, Let us consider timber-plate structure defined by its length L
technical approvals and very active research. The design codes, and width B, see Fig. 1. Let the structure consists of: (i) CLT plate
e.g. EC5 [2], do not consider CLT material. From the list of EC5 con- and (ii) nr timber ribs with rectangular cross-sections. Let us
sidered materials, the glued laminated timber (GLT) is the most assume that the ribs are parallel, evenly arranged and of the same
similar, thus the GLT design values and rules are commonly cross-section. The ribs are arranged along the width as shown in
adopted for CLT design. Fig. 1. Consider L and B given. The other structural dimensions,
Due to long tradition of 1d elements in timber constructions, thicknesses of CLT plate layers (e.g. t1 ; t 2 ; t3 for a 3 layer CLT
many CLT analyses are based on beam theories. Such 1d theories plate), ribs width br and height hr , as well as the number of ribs
can produce results that considerably deviate from experimentally nr , are the design variables that are to be computed by
measured values, see e.g. [3]. It has become apparent that plate optimization.
theories should be applied for CLT analysis, delivering more reli- The following optimization problem is considered: find such
able deformation and stress states, e.g. [4–6]. Experimental and values of the design variables that: (i) the structural volume is min-
computational data on realistic CLT material parameters is also imal, and (ii) the design requirements (defined by EC5) are fulfilled.
increasing, e.g. [1,7,8]. In the mathematical terms: find vector / that will minimise scalar
Despite of the considerable growth of CLT in the world market, function Uð/Þ, which is defined as a weighted sum of the objective
designers are still far from exploiting its maximum potential. An function U1 ð/Þ ¼ Vð/Þ and nU  1 constraint functions
Ui ; i ¼ 2; . . . ; nU . More precisely, find
⇑ Corresponding author.
minðUð/ÞÞ; ð1Þ
E-mail address: bbrank@fgg.uni-lj.si (B. Brank).

http://dx.doi.org/10.1016/j.compstruct.2016.07.072
0263-8223/Ó 2016 Elsevier Ltd. All rights reserved.
528 A. Stanić et al. / Composite Structures 154 (2016) 527–537

Fig. 1. Rectangular timber-plate structure with nr ¼ 2 ribs.

where applicable if the constraint functions are designed as penalty func-


X
nU X
nU tions. The constraint functions are constructed with Bð/L ; /R ; l; eÞ,
Uð/Þ ¼ wi Ui ð/Þ ¼ w1 Vð/Þ þ wi Ui ð/Þ ð2Þ which plays the role of the penalty function, see Table 1. Due to
i¼1 i¼2 B, the U returns a large value at a constraint violation, which
directs the optimization towards a change compliant with the vio-
/ ¼ f/1 ; /2 ; /3 ; . . . ; /n/ gT consists of n/ design variables, bounded as
lated constraint.
/min
i 6 /i 6 /max
i ; wi are weights, and V is structural volume. The Function B is defined as
constraint functions Ui ; i ¼ 2; . . . ; nU and related constraints are (
 lnð/L  /R Þ if /L  /R > e;
summarized in Table 1. Bð/L ; /R ; l; eÞ ¼ l /2R 2/R /L þ/2L þ4/R e4/L eþ3e2 2e2 lnðeÞ
We choose to analyse structural response of the timber plate by 2 e2
else
the finite element method, in particular by solid element assessed ð4Þ
in Section 3. According to EC5 [2], structural response has to be
checked for various load cases, Table 1: the serviceability limit where l and e are small numbers, the weight factor and tolerance,
state (SLS) load case, the ultimate limit state (ULS) load case, the respectively. Suppose, that B is used to enforce /L > /R . When
fundamental eigenfrequency load case (xLS), the vibration check /L > /R þ e is true, Bð/L ; /R ; l; eÞ returns a small number. If the
load case (vLS), and the fire resistance load case (FLS). inequality is not satisfied, a large number is returned. An example
In Table 1, B denotes a function explained in Section 2.2. The of penalty function Bð/L ; 0:1; 104 ; 104 Þ, related to /L > 0:1, is plot-
indices i; j; k relate to the design variable, mesh node, and stress ted on Fig. 2; for /L > 0:1 its value is practically zero, and for
vector component, respectively. Other notations are: nn is the /L < 0:1 it increases sharply. Similarly, the constraint /R < 0:9 can
number of nodes of the mesh, nn;S is the number of nodes of the be controlled by Bð0:9; /R ; 104 ; 104 Þ, Fig. 2.
j
mesh top surface, nr is the number of stresses that are to be
checked at the jth node of the mesh, uz is displacement in z direc- 3. Finite element used
tion, x1 is the fundamental eigenfrequency, uz;d is the maximal
allowed displacement, xd is the minimal allowed fundamental fre- 3.1. Basic features of the element
quency and r  are normalized stresses. For node j, the sets r
 ULS
j and
r j are (a 2 fULS; FLSg)
FLS In order to compute the displacements, stresses and eigenfre-
quency that take part in the constraint functions one needs to per-
r aj ¼ fr aj;k jk ¼ 1; . . . ; nrj g ð3Þ form finite element analyses at each optimization iteration. We
model the structure as a solid, which allows for detailed analysis

2.2. Optimization concept

Optimization is based on minimization of weighted sum of the


objective and constraint functions, see Eq. (2). This concept is

Table 1
Constraints and constraint functions.

Constraint Constraint function


Pn/  
/min
i 6 /i U2 ð/Þ ¼ i¼1 B /i =/min i ; 1; l; e
Pn/  
/i 6 /max
i U3 ð/Þ ¼ i¼1 B 1; /i =/maxi ; l; e
j uSLS Pnn;S  SLS 
z;j j 6 uz;d U4 ð/Þ ¼ j¼1 B j uz;j j =uz;d ; 1; l; e
SLS SLS
  2 
jr
 ULS
j;k j6 1
P n Pnrj
U5 ð/Þ ¼ nj¼1 k¼1
B 1; r  ULS
j;k ; l; e
 
xx P xx U6 ð/Þ ¼ B xx =xx d ; 1; l; e
LS LS LS LS
1 d
 1 
j uvz LS j 6 uvz;dLS v v
U7 ð/Þ ¼ B 1; j uz j =uz;d ; l; e
LS LS
  2 
jr
 FLS
j;k j6 1
P n Pnrj
U8 ð/Þ ¼ nj¼1 k¼1
B 1; r  FLS
j;k ; l; e
Fig. 2. Enforcement of two constraints by B.
A. Stanić et al. / Composite Structures 154 (2016) 527–537 529

of the CLT-rib connection. For this purpose one needs a solid or with h ¼ 0 . Regarding (7) and (8), we note that it is common in CLT
solid-shell element that: (i) is reliable for thin plate problems (no analysis not to distinguish between T and R [4].
shear locking), (ii) delivers reasonable approximations for the For EAS elements there is no unique way to compute the ele-
transverse shear stresses in laminated composite, (iii) captures ment stresses [10]. In this work, we compute them from W. Vector
well the ‘‘in-plane” bending, and (iv) is computationally fast of 2nd Piola–Kirchhoff stresses is
(allows using relatively fine meshes).
@W
The solid element that complies reasonable well with the above S¼ ¼ ðTCTT ÞE ð9Þ
@E
demands is presented in [9]. The element has 8 nodes, 12 enhanced
assumed strain (EAS) modes [10] and 8 Gauss integration points. and vector of Cauchy stresses r ¼ frxx ; ryy ; rzz ; ryz ; rxz ; rxy gT is
The basic assumption is decomposition of the displacement gradi- available after transformation r ¼ FSFT = det F, where r and S are
ent into the standard (displacement) part Hu and an enhanced part matrices with Cauchy and 2nd Piola–Kirchhoff stresses, respec-
e so that the deformation gradient is F ¼ Hu þ H
H, e þ I. The element tively. The Cauchy stresses in material coordinates are
does not exhibit shear locking for isotropic thin plate and
‘‘in-plane” problems [9].
rmat ¼ T1 r ð10Þ
The element was modified for ribbed CLT plate. The strain where r mat
¼ fr11 ; r22 ; r33 ; r23 ; r13 ; r12 g .
T

energy function for orthotropic material was implemented Once r is known at element Gauss points, the ‘‘in-plane” stres-
1  ses rxx ; ryy and rxy are extrapolated to the element nodes. Since
W¼ ðTCTT ÞE  E ð5Þ the ‘‘in-plane” stresses are discontinuous in laminated composites,
2
we do not smooth them across the mesh. However, we smooth
where E ¼ fexx ; eyy ; ezz ; 2eyz ; 2exz ; 2exy gT is vector of the components rxz ; ryz and rzz (e.g. [12]) as they are in reality smooth through
of the Green–Lagrange strain matrix E defined as E ¼ 12 ðFT F  IÞ, the laminated composite plate thickness.
and C defines the stress–strain relations in material axes, e.g. [11] The derived element is geometrically nonlinear. One can use it
2 31 as nonlinear (e.g. for buckling check) but in most cases its linear
1
E
 mE212  mE313 0 0 0 version suits perfectly well the timber plate problems. To illustrate
6 m1 7
6  12
6 E1
1
E2
 mE323 0 0 0 7
7
its performance, results of two linear tests are presented below.
6 m13 7
6
6 E1  mE232 1
0 0 0 7
7
C¼6
E3
7 ð6Þ 3.2. Laminated beam test
6 0 0 0 1
0 0 7
6 G23 7
6 7 We consider three-layered simply supported beam with
6 0 0 0 0 1
0 7
4 G13 5 L ¼ 6 m, width B ¼ 1 m and t=L ¼ 0:1, Fig. 4. It is supported at its
0 0 0 0 0 1  t 3
G12 the middle axis and loaded by p ¼ 0:1L p0 at the top surface;
In (6), E1 ; E2 and E3 are the elasticity moduli in the longitudinal L, p0 ¼ 5  105 N/m2. It is composed of two isotropic materials:
transverse T and radial R directions of the timber lamina, denoted top and bottom layers have E ¼ 11  109 N=m2 ; m ¼ 0;
as material axes 1 ¼ L; 2 ¼ T and 3 ¼ R, respectively, Fig. 3. The G ¼ 5:5  10 N/m , and the middle layer has E ¼ 0:5  109 N=m2 ;
9 2

shear moduli in 12, 13 and 23 planes are G12 ; G13 and G23 , respec-
m ¼ 0; G ¼ 0:25  109 N/m2. As for the mesh, the span is divided
tively, and m12 ¼ m21 EE12 ; m13 ¼ m31 EE13 and m23 ¼ m32 EE23 are the Poisson’s into nL , the width into nB , and the height into nt elements.
rations (where mij ¼  eejjii ). Results for the meshes nL  nB  nt 2 f8  2  6; 24  4
The matrix T in (5) represents transformation between 1; 2; 3 6; 48  8  6g are presented in Figs. 5 and 6. They are compared
and x; y; z. For CLT, see Fig. 1, axis 3 ¼ z, and T for the Gauss points to the analytical solution for the 3-layered Euler–Bernoulli beam.
in CLT part of the mesh is [11] The computed rxx coincides with the analytical solution, and the
2 3 approximation for rxz is reasonable.
2
cos2 ðhÞ sin ðhÞ 0 0 0  sinð2hÞ
6 7
6 2
sin ðhÞ cos2 ðhÞ 0 0 0 sinð2hÞ 7 3.3. Laminated plate test
6 7
6 7
6 0 0 1 0 0 0 7
T ¼ Txy ¼ 6 7
6 0 0 0 cosðhÞ sinðhÞ 0 7 We consider 3-layered square plate (Fig. 7), which has span
6 7
6 7 a ¼ 20, thickness t, the 0 =90 =0 layers are h ¼ t=3 thick, and the
4 0 0 0 sinðhÞ cosðhÞ 0 5
sinðhÞ cosðhÞ sinðhÞ cosðhÞ 0 0 0
2
cos2 ðhÞ  sin ðhÞ following supports
ð7Þ uy ð0; y; zÞ ¼ uy ða; y; zÞ ¼ 0; uz ð0; y; zÞ ¼ uz ða; y; zÞ ¼ 0
ð11Þ
where h is the angle in xy parallel plane that represents rotation of x ux ðx; 0; zÞ ¼ ux ðx; a; zÞ ¼ 0; uz ðx; 0; zÞ ¼ uz ðx; a; zÞ ¼ 0
into axis 1 around þz. The matrix (7) is also used for the ribs part of    
The top surface is loaded by qðx; yÞ ¼ q0 sin pax sin pay . The material
the structure, with h ¼ 0 , if the ribs are as in Fig. 3 (middle). If the
ribs are as in Fig. 3 (right), the transformation matrix for the Gauss data are: E1 ¼ 25  106 ; E2 ¼ E3 ¼ 106 ; m12 ¼ m13 ¼ m23 ¼ 0:25;
points in the ribs part of the mesh is G12 ¼ G13 ¼ 0:5  106 , and G23 ¼ 0:2  106 . Two plates with

2 2
3
cos2 ðhÞ sin ðhÞ 0 0 0  sinð2hÞ
6 7
6 0 0 1 0 0 0 7
6 7
6 sin2 ðhÞ cos2 ðhÞ 0 0 0 sinð2hÞ 7
6 7
T ¼ Txz ¼ 6 7 ð8Þ
6 0 0 0 cosðhÞ sinðhÞ 0 7
6 7
6 2 7
4 sinðhÞ cosðhÞ  sinðhÞ cosðhÞ 0 0 0 cos ðhÞ  sin ðhÞ 5
2

0 0 0  sinðhÞ cosðhÞ 0
530 A. Stanić et al. / Composite Structures 154 (2016) 527–537

Fig. 3. Local axes in CLT layer (left) and timber rib (middle and right).

Fig. 4. Laminated beam.

Fig. 5. Beam: rxx at x ¼ L=2.

Fig. 6. Beam: rxz at x ¼ 0.

   
t ¼ 0:01a and t ¼ 0:25a are analysed. The meshes are of E2 100E2
u¼ ux ; w ¼ uz
nx  ny  nz 2 f8  8  6; 12  12  6g elements (Fig. 8), where nx ; q0 tr 3
q0 tr 4

ny and nz are numbers of elements in x; y and z directions, respec-   ð12Þ


1 a
tively. The computed displacements and stresses are normalized as frxz ; ryz g ¼ frxz ; ryz g; r ¼
q0 r t
A. Stanić et al. / Composite Structures 154 (2016) 527–537 531

Fig. 7. Plate: geometry and loading.

Fig. 8. Plate: 12  12  6 meshes.

and compared to analytical solution [13]. The results for u are pre- The EC5 [2] defines the load-case-dependent design moduli and
sented on Fig. 9 (left). For the point ðx; y; zÞ ¼ ða=2; a=2; t=2Þ, the strengths. For ULS; SLS; xLS and vLS, the design moduli are
computed displacements for thin plate are w886 ¼ 0:417 and ðÞd ¼ ðcÞk and the design strengths are ðÞd ¼ kmod ðcÞk , where ðÞk
w12126 ¼ 0:420, while w ¼ 0:433 [13]. For thick plate one has
M M

denotes the characteristic value, kmod is the modification factor


w886 ¼ 2:100 and w12126 ¼ 2:093, and w ¼ 2:118 [13]. Compar-
and cM is the safety factor. For FLS, the design moduli are
ison of computed and analytical transverse shear stresses is shown kfi ðÞ0:05
in Figs. 9 (right) and 10. The agreement is not perfect, but it is ðÞd;fi ¼ kmod;fi cM;fi and the design strengths are ðÞd;fi ¼
kfi ðÞk
acceptable. kmod;fi cM;fi , where kmod;fi is the modification factor, cM;fi is the safety
factor and ðÞ0:05 denotes the 5% fractile of the modulus at room
4. Considered load cases temperature.

In this section, we define the load cases relying on Eurocodes. 4.1. Ultimate limit state (ULS)
We consider the self-weight, characteristic permanent action g k ,
and characteristic live load ql . They are applied as uniform surface Let us assume the medium-term load-duration class [2]; in that
loads on the top surface of the mesh. The surface load due to self- case kmod ¼ 0:8 and cM ¼ 1:25. The stresses computed for the ULS
weight is g l ¼ ðqCLT V CLT þ qribs V ribs Þ=A, where q is material density, design load qd;ULS ¼ 1:35ðg l þ g k Þ þ 1:5qk should not exceed the
V ¼ V CLT þ V ribs is structural volume (recomputed at every opti- design strengths. Table 2 and Fig. 11 illustrate how the stresses
mization step) and A is plate area. are checked. We assume that the rib is glued to the 1st and 2nd

Fig. 9. Plate t ¼ 0:25a. Left: u at ðx; yÞ ¼ ð0; a=2Þ. Right: plate (t ¼ 0:01a): rxz at ðx; yÞ ¼ ð0; a=2Þ.
532 A. Stanić et al. / Composite Structures 154 (2016) 527–537

Fig. 10. Plate t ¼ 0:25a. Left: rxz at ðx; yÞ ¼ ð0; a=2Þ. Right: ryz at ðx; yÞ ¼ ða=2; 0Þ.

Table 2 and the FLS design strengths are calculated with kmod;fi ¼ 1; cM;fi ¼ 1,
Stress control (see Fig. 11).
and the coefficient for GLT kfi ¼ 1:15, see [14]. The plate with the
Control zone r
 aj ; a 2 fULS; FLSg
reduced dimensions is loaded by the accidental load combination

1, 2, 3 ra ra jra j jra j for fire qd;FLS ¼ g l þ g k þ 0:30qk .
tens:=comp: : f a 11 ; f a 22 ; shear : f12=3
a ; f a 23
t=c;0;d t=c;90;d V;d V;R;d

4 jra11 j
a
f m;d
4.3. Serviceability limit state (SLS)

5 r
j a11 j r
j a12 j
;
f am;d f aV;d The SLSload case is used to check the maximum deflection. The

6 jra12 j design material properties are obtained with cM ¼ 1. The maximal
a
f V;d
computed deflection for qd;SLS ¼ g l þ qk is checked against the
minðB;LÞ
z;d ¼
allowed one uSLS 300
.
layers of CLT (numbered from the bottom); a simplified version
4.4. Load case for modal analysis (xLS)
exists with rib glued to the bottom face of CLT.

An important control of timber plates refers to vibrations. EC5


4.2. Fire resistance load case (FLS)
[2] requires that the fundamental frequency of the timber plate
is higher than xx d
LS
¼ 8 Hz. The self-weight and the permanent
For the FLS, the effective cross-section is defined. Let us assume
load qk are used to determine mass matrix M, and the fundamental
that the fire exposure time is t ¼ 30 min [14]; the effective char-
frequency x1 is computed from the generalized eigenvalue prob-
ring depth def for timber with unprotected surfaces is then
lem K  x21 M ¼ 0, where K is the structural stiffness matrix
def ¼ bn t þ k0 d0 ¼ 0:7 mm=min  30 min þ1  7 mm ¼ 28 mm (cM ¼ 1).
ð13Þ
4.5. Serviceability limit state for vibration (vLS)
where the values of bn (design notional charring rate for GLT), k0
and d0 are given in (13). The effective cross-section is obtained by Another vibration related control from [2] requires that the tim-
reducing the cross-section by def from the bottom side (Fig. 12) if ber plate-structure is loaded by the vertical force F d;v LS ¼ 1 kN at
the fire is assumed to be underneath the plate. The FLS material data the ‘‘worst-possible” location and that the computed maximal

t1 Control zone 1
t2 t
Control zone 2
Hr
3 Control zone 3

Hr
hr Control zone 4
3
z
Hr Control zone 5
3
y Control zone 6
br

Fig. 11. Stress control zones.


A. Stanić et al. / Composite Structures 154 (2016) 527–537 533

t1 t1
t2 t t2 tred
t3 t3,red

hr def hr,red
z z

y y
br,red
br

Fig. 12. Cross-section for ULS (left) and effective cross-section for FLS (right).

deflection is less than uvz;dLS ¼ 1 mm. We apply F d;v LS at the top The gradient-based optimization method will find any opti-
surface node with the maximal SLS deflection from the previous mum, either local or global. In this work no special algorithm has
optimization iteration. For this load case, cM ¼ 1. been applied that would be helpful in finding the global optimum.
The most obvious ‘remedy’ to that problem is to start the optimiza-
5. Optimisation procedure tion process from various scattered points of the design space. This
can improve the probability to compute the global optimum (or at
The gradient method, see e.g. [15], has been used to solve the least a close local one).
above defined optimization problem. For the considered structural
problem, the finite element analyses (i.e. the structural response 5.2. Derivatives for sensitivity analysis
computations) are quite costly computationally, especially when
one chooses to use the geometrically nonlinear version of the The basic ingredient of the structural shape optimization is con-
above described finite element. On the contrary, the sensitivity nection between the mesh and the design variables /. Since we
analysis is computationally cheap, because the number of design deal with simple structural geometry, this connection is easily
variables is low and the analytical expressions were derived for established and does not require usage of e.g. design body concept
the derivatives by using a direct differentiation approach [16]. [18,19]. The coordinates can be straightforwardly described with
dXð/Þ
Thus, the preferred optimization method is the one with the lowest the design variables Xð/Þ and the derivative d/
can be easily
number of needed structural response computations. In this view, obtained.
the gradient-based methods are clear favourites, since they For the gradient optimization method, the gradient of U with
demand smaller number of finite element analyses as the respect to the design variables must be evaluated
stochastic methods (like genetic algorithms).
@U X
n
U
@ Ui ð/Þ
¼ wi ð14Þ
5.1. Procedure outline @/ i¼1
@/

The applied optimization consists of the following: The evaluation of dUd/


i ð/Þ
is straightforward for i 2 f1; 2; 3g. The evalu-
ation of (14) for i 2 f4; 5; 7; 8g is more demanding, since it includes
@u
1. Set the iteration counter to l ¼ 0. Choose the initial design /ð0Þ . knowledge of @/ , where vector u collects non-prescribed nodal dis-
2. Perform the finite element analyses: compute nodal displace- placements of the mesh. The needed derivative can be obtained
ments uðlÞ for SLS and vLS, stresses rðlÞ for ULS and FLS and fun- from the equilibrium equation
damental frequency x1 for xLS. The code AceFEM [16] is used
ðlÞ
Rint ðuð/Þ; /Þ ¼ Rext ð/Þ ð15Þ
in this step.
int ext
3. Compute objective and constraint functions listed in Table 1 by where R and R are vectors of the internal and external nodal
using the results of the finite element analyses. forces, respectively. Derivation of (15) with respect to / leads to
 ðlÞ
4. Perform sensitivity analysis: compute @@/U that includes com- @Rint @u @Rint dRext
 ðlÞ  ðlÞ  ðlÞ  ðlÞ þ ¼ ð16Þ
x1 @u @/
|ffl{zffl} @/ d/
putation of derivatives dV d/
@u
; @/ ; @@/r and dd/ . The
K
AceGen [16] symbolic code generator is used to derive analytical
@u
expressions for these derivatives [9]. where K is the (tangent) stiffness matrix. One can compute @/ from
5. Compute an update of design parameters D/ðlÞ by gradient opti- (16) as
!
mization method. The gradient based optimization procedure @u dRext @Rint
implemented in Mathematica’s [17] FindMinimum function is ¼ K1  ð17Þ
@/ d/ @/
used.
6. Define new design by /ðlþ1Þ ¼ /ðlÞ þ D/ðlÞ . In the finite element code, Rint ; Rext and their derivatives are
7. Check for the convergence: if it is fulfilled, exit the optimization, assembled on the element-by-element basis, i.e.
otherwise set l ¼ l þ 1 and go to 2. The iterations are repeated
Ne Ne
until Uð/Þ changes negligibly for a number of subsequent Rint ¼ A Rint;ðeÞ ; Rext ¼ A Rext;ðeÞ ; ð18Þ
iterations. e¼1 e¼1
534 A. Stanić et al. / Composite Structures 154 (2016) 527–537

@r
@Rint Ne @Rint;ðeÞ dRext Ne dRext;ðeÞ and j
@/ ðeÞ
at the element Gauss points and extrapolate those values
¼ A ; ¼ A ð19Þ
@/ e¼1 @/ d/ e¼1 d/ to the element nodes. The derivatives at a mesh node with respect
@u @r
to / can be taken from @/ and ð@/ ÞjðeÞ .
where A is the finite element assembly operator and Ne is the num-
dUi
ber of elements in the mesh. The element contribution to the vector The analytical expressions for d/
; i – 6, are thus available. On
of internal forces is the other hand, the U6 entry x1 is not available as x1 ð/Þ. The finite
Z
difference is used to approximate its derivative
Rint;ðeÞ ¼ BðeÞ;T S ðeÞ det JðeÞ dndgdf ð20Þ
 ðlÞ

dx1 1   ðlÞ   
ðeÞ @E
¼ x 1 /  x 1 /ðlÞ
 D /ðl1Þ
ð23Þ
where  is the bi-unit cube, B ¼ j
¼ BðXð/Þ; uð/ÞÞjðeÞ is the ele-
@u ðeÞ d/ D/ðl1Þ
ment strain–displacement matrix (which depends on u for geomet-
 
rical nonlinearity), S ðeÞ ¼ TCTT E jðeÞ ¼ SðXð/Þ; uð/ÞÞjðeÞ is the 6. Examples
element stress vector, and JðeÞ ¼ @X j ¼ JðXð/ÞÞjðeÞ is the element
@n ðeÞ
Jacobian matrix related to the element mapping from We present two numerical examples. We assume the timber
quality class C24, which is mainly used in commercial production.
X ¼ fx; y; zg to n ¼ fn; g; fg . The derivative
T T @Rint;ðeÞ
@/
is
The characteristic material parameters are given in Table 3
Z !
int;ðeÞ T
@R @B dX @S dX dðdetJÞ dX
¼ S detJ þ BT detJ þ BT S dndgdf
@/  @X d/ @X d/ dX d/
jðeÞ Table 5
Example 1: supports (see Fig. 13).
ð21Þ
Edge Displacement constraints
dRext;ðeÞ
The derivative d/
can be obtained in a more straightforward
ðx; zÞ ¼ ð0; 0Þ ux ¼ 0; uz ¼0
manner, since the only design-variables-dependent loading is the ðx; zÞ ¼ ðL; 0Þ ux ¼ 0; uz ¼0
self weight. ðy; zÞ ¼ ð0; 0Þ uy ¼ 0; uz ¼0
@u
Once @/ is obtained from (17), one can compute ðy; zÞ ¼ ðB; 0Þ uy ¼ 0; uz ¼0
ðx; zÞ ¼ ð0; hr Þ ux ¼ 0; uy ¼ 0; uz ¼ 0
      ¼ 0; ¼ 0; uz ¼ 0
@S @S @u @S @E ðx; zÞ ¼ ðL; hr Þ ux uy
j ¼ ; j ¼ TCTT j ¼ TCTT B jðeÞ ð22Þ
@/ ðeÞ @u @/ jðeÞ @u ðeÞ @u ðeÞ

Table 3
Characteristic material parameters.
4
Elastic moduli Shear moduli [GPa] Poisson
[GPa] coefficients
E1 11 G12 0.688 m12 0.09
E2 0.37 G13 0.688 m13 0.36 3
E3 0.37 G23 0.050 m23 0.38

2
Table 4
Characteristic strengths.

Strength [MPa]
1
Compression f c;0;k and f c;90;k 24.0 and 3.0
Tension f t;0;k and f t;90;k 14.0 and 0.5
Bending f m;k 24.0
Step
In-plane shear f V;k 2.5 0 50 100 150 200
Rolling shear f V;R;k 0.8
Fig. 14. Example 1: Uð/Þ during optimization.

y
B=6m

z
hr

br er = 2 m

x
B=6m
L = 2.95 m

Fig. 13. Example 1: geometry.


A. Stanić et al. / Composite Structures 154 (2016) 527–537 535

Table 6 in the Table 4 (the values are from [20], except for f c;90;k ; f t;90;k and
Example 1: results in cm (V is in m3 ).
f V ;R;k that are from [1]). The material density is q ¼ 420 kg/m3 [1].
/i /min
i
/max
i Initial value Optimized value We choose g k ¼ 0:25 kN/m2 and qk ¼ 2 kN/m2 as defined in [22,23]
t1 1.5 4.5 3 1.50 for plates in residential building.
t2 1.5 4.5 3 1.50
t3 1.5 4.5 3 1.50 6.1. Example 1: CLT plate with 3 ribs
br 2.5 20 15 3.49
hr 6 40 20 38.35
We consider plate-structure with L ¼ 2:95 m, B ¼ 6 m (Fig. 13)
V 1.86 0.91 and three ribs. The CLT has 3 layers with thicknesses t 1 ; t2 and t3
(counted from the top) and 0 =90 =0 orientations. For boundary
conditions see Fig. 13 and Table 5. The design variables are thick-
nesses of the layers and height and width of the ribs. They are nor-
malized by hr initial value in order to be numerically comparable.
The weights in Eq. (2) are set to 1.
Fig. 14 shows how Uð/Þ changes during the optimization. Sharp
function increases are noticed at constraint violations. The results
are presented in Table 6. The optimized CLT thicknesses are the
Fig. 15. Example 1: optimal configuration. minimal allowed. The optimized height of the ribs is quite large
due to its important contribution to the bending stiffness. Fig. 15
shows optimized configuration. The active constraint functions
were U2 ð/Þ; U4 ð/Þ and U5 ð/Þ.

6.2. Example 2: ribbed CLT plate with hinges

We consider the plate-structure with B ¼ 10 m and L ¼ 7 m


(Figs. 16 and 17). The CLT plate has three layers with thicknesses
t1 ; t 2 and t3 (numbered from the top). The structure is composed
of 4 modular units that are connected by hinges, modelled as
shown in Fig. 18 (D/ in Fig. 18 is 5 mm). Each modular unit with
B1 ¼ 2:5 m and L ¼ 7 m has 3 ribs, which are arranged evenly along
B. The axial distance between the ribs is 10=12 m. The supports are
presented in Table 7.
The weights in (2) are w1 ¼ 10 and wi;i–1 ¼ 100. The design vari-
ables t 1 ; t2 ; t3 and br are normalized by the initial value of hr . The
optimisation was run for 9 different sets of the initial values of
design variables, which produced different optimized configura-
tions (among them 6 have very similar volume (from 4.71 to
4.81 m3), another 2 have slightly bigger volume (more than
5 m3), while one solution is out of the scope with the volume more
than 10 m3), see Table 8, where init and opt are for the initial and
optimized values, respectively. Fig. 19 shows evolution of Uð/Þ;
it has a decreasing trend when constraints are not violated. The
active constraint functions are always U5 ð/Þ and U8 ð/Þ. Fig. 20
shows optimized value with respect to number of ribs per modular
unit.

7. Conclusions

Fig. 16. Example 2: geometry. A robust optimization procedure for a civil engineering prob-
lem, which can be stated as ‘‘find an economic design of CLT plate
(with stiffening ribs) that complies with the EC5 timber code
(E1 ; E2 ; E3 ; G12 and G13 values are from [20], the value for G23 is demands” has been presented. The proposed procedure is based
from [4] and the values of Poisson’s coefficients are computed from on the use of penalty function, which allows for an easy inclusion
CLT experimental data [21]). The characteristic strengths are listed of different constraints. For example, an inclusion of the dynamic

Fig. 17. Example 2: cross-section.


536 A. Stanić et al. / Composite Structures 154 (2016) 527–537

Fig. 18. Example 2: modeling of the hinge in CLT.

Table 7
Example 2: supports (see Fig. 16).

Surface/edge Prescribed displacement 100

x¼0 uz ¼0
x¼L uz ¼0 80
ðx; zÞ ¼ ð0; 0Þ ux ¼0
ðy; zÞ ¼ ð0; 0Þ uy ¼0
ðy; zÞ ¼ ðB; 0Þ uz ¼0 60

40
vibrational response of the plate-structure into the optimization
could be handled by yet another constraint function (and the 20
dynamic finite element analysis).
The finite element analyses were performed by modified solid
Step
element from [9] that produces reasonable through-the-thickness 0 20 40 60 80 100
transverse shear stresses distributions and accurate through-the-
thickness distributions of displacements and in-plane stresses. In Fig. 19. Example 2: objective function Uð/Þ during the optimization.
further studies, one might consider improving its capabilities or
replace it with a solid or solid-shell element designed for lami-
nated composites, e.g. [24].
For the optimization problem discussed, the gradient-based
methods [15] are more efficient than other, since they require
smaller number of finite element computations of structural
response. Such an optimization problem may be solved in its orig-
inal form (by handling the constraints as they are, see e.g. [18,19])
or by packing all the constraints into a modified objective function
(with a suitable penalty function), as is the case in the present
work. Here, a general advice what is better is hardly available
and the choice simply depends on the problem at hand. We used Fig. 20. Example 2: optimized volume with respect to number of ribs.
the latter option because it helped us to define and run the opti-
mization problem.
We note that due to the 3d finite element modelling the compu- plate. However, they are not extremely large even if one uses the
tational times can be relatively large for a fine mesh and large nonlinear version of the finite element.

Table 8
Example 2: summary of optimisation runs. The one with the most economic final design is bold.

N ti;init br;init hr;init V init t1;opt t2;opt t 3;opt br;opt hr;opt V opt

1 4.4 19 39 15.46 1.50 1.51 1.50 7.75 24.39 4.75


2 3.0 10 30 8.82 1.50 1.50 1.50 7.47 24.87 4.71
3 4.0 19 10 10.00 4.07 4.08 4.07 19.04 10.91 10.30
4 1.5 8 35 8.82 1.50 1.50 1.50 7.50 24.83 4.72
5 3.5 19 19 10.38 1.50 1.51 1.50 17.55 22.36 5.45
6 2.5 10 30 7.77 1.50 1.50 1.50 7.51 24.80 4.72
7 2.0 19 30 8.99 1.51 1.51 1.52 8.00 23.97 4.79
8 3.0 8 20 7.64 1.63 1.60 1.83 8.33 23.11 5.17
9 2.4 11 31 7.90 1.50 1.50 1.50 8.47 23.29 4.81
A. Stanić et al. / Composite Structures 154 (2016) 527–537 537

We believe that there are several novel aspects of CLT plate [8] Flores E, Saavedra K, Hinojosa J, Chandra Y, Das R. Multi-scale modelling of
rolling shear failure in cross-laminated timber structures by homogenisation
design presented in this work. Since the use of CLT in timber build-
and cohesive zone models. Int J Solids Struct 2016;81:219–32.
ing constructions is increasing rapidly, an economic-design optimi- [9] Korelc J, Šolinc U, Wriggers P. An improved EAS brick element for finite
sation can have a significant practical importance. deformation. Comput Mech 2010;46(4):641–59.
[10] Simo J, Rifai M. Class of mixed assumed strain methods and the method of
incompatible modes. Int J Numer Meth Eng 1990;29(8):1595–638.
Acknowledgement [11] Reddy JN. Mechanics of laminated composite plates and shells: theory and
analysis. CRC press; 2004.
[12] Bohinc U, Brank B, Ibrahimbegovic A. Discretization error for the discrete
The financial support of the WoodWisdom-Net+ initiative and Kirchhoff plate finite element approximation. Comput Methods Appl Mech
the Slovenian Ministry of Education, Science and Sport is gratefully Eng 2014;269:415–36.
acknowledged. We thank Mr. Iztok Šušteršič for helpful discussions [13] Pagano NJ. Exact solutions for rectangular bidirectional composites and
sandwich plates. J Compos Mater 1970;4:20–34.
on CLT design and prof. Marko Kegl for helpful discussions on [14] EN 1995-1-2: Eurocode 5: Design of timber structures – part 1-2: general –
gradient optimization. structural fire design; 2004.
[15] Kegl M, Butinar BJ, Kegl B. An efficient gradient-based algorithm for
mechanical systems. Commun Numer Methods Eng 2002;18:363–71.
References [16] J. Korelc, AceGen and AceFEM user manual, Available at <http://symech.fgg.
uni-lj.si/>. University of Ljubljana; 2015.
[1] Brandner R, Flatscher G, Ringhofer A, Schickhofer G, Thiel A. Cross laminated [17] W.R. Inc., Mathematica version 10, Wolfram Research Inc., Champaign, Illinois;
timber (CLT): overview and development. Eur J Wood Wood Prod 2016:1–21. 2016.
[2] EN 1995-1-1: Eurocode 5: Design of timber structures – part 1-1: General – [18] Kegl M, Brank B. Shape optimization of truss-stiffened shell structures with
common rules and rules for buildings; 2004. variable thickness. Comput Methods Appl Mech Eng 2006;195(19-
[3] Czaderski C, Steiger R, Howald M, Olia S, Gülzow A, Niemz P. Tests and 22):2611–34.
calculations on 3-layered cross-laminated solid wood panels supported at all [19] Kegl M, Brank B, Harl B, Oblak M. Efficient handling of stability problems in
edges [versuche und berechnungen an allseitig gelagerten 3-schichtigen shell optimization by asymmetric ‘worst-case’ shape imperfection. Int J Numer
brettsperrholzplatten]. Holz als Roh – und Werkstoff 2007;65(5):383–402. Meth Eng 2008;73(9):1197–216.
[4] Stürzenbecher R, Hofstetter K, Eberhardsteiner J. Structural design of cross [20] EN 338: Structural timber – strength classes; 2003.
laminated timber (CLT) by advanced plate theories. Compos Sci Technol [21] Gsell D, Feltrin G, Schubert S, Steiger R, Motavalli M. Cross-laminated timber
2010;70(9):1368–79. plates: evaluation and verification of homogenized elastic properties. J Struct
[5] Brank B. On composite shell models with a piecewise linear warping function. Eng 2007;133(1):132–8.
Compos Struct 2003;59(2):163–71. [22] EN 1990: Eurocode: basis of structural design; 2002.
[6] Kreja I. A literature review on computational models for laminated composite [23] EN 1991-1-1: Eurocode 1: actions on structures – part 1-1: general actions –
and sandwich panels. Cent Eur J Eng 2011;1(1):59–80. densities, self-weight, imposed loads for buildings; 2000.
[7] Zhou Q, Gong M, Chui Y, Mohammad M. Measurement of rolling shear [24] Vu-Quoc L, Tan X. Efficient hybrid-EAS solid element for accurate stress
modulus and strength of cross laminated timber fabricated with black spruce. prediction in thick laminated beams, plates, and shells. Comput Methods Appl
Constr Build Mater 2014;64:379–86. Mech Eng 2013;253:337–55.

You might also like